Fall Detection in the Elderly using Different Machine Learning Algorithms with Optimal Window Size

Mobility deteriorates with age, and falls become more frequent, resulting in injuries or even death. However, many such injuries and deaths can be prevented, resulting in financial savings. This paper proposes a method to identify and distinguish falls from activities of daily living (ADLs) in older adults utilizing a wearable fall detection device. Novel preprocessing and feature extraction techniques to extract features from accelerometry data were developed. In addition, machine-learning techniques, such as support vector machine (SVM), k-nearest neighbor (k-NN), random forest (RF), and artificial neural network (ANN), were used to classify the acceleration and rotation signals into falls or ADLs. The main contributions of the research are the implementation of a two-class classification algorithm for fall detection and analyzing the effect of sliding window size on system performance. The publicly available SisFall dataset was utilized to develop the fall detection algorithm, and various window sizes were evaluated. The results show that the best compromise between processing time and detection performance is achieved with a window size of 3 s. The proposed and implemented approaches employing the SVM algorithm demonstrated a perfect F-1 score and recall value of 100% when testing for a fall. We achieved an accuracy of 96.34% by using the k-NN algorithm. Furthermore, the ensemble machine-learning algorithms, SVM and RF, achieved accuracy, sensitivity, and specificity higher than 99%.